[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86395-en":3,"doc-seo-86395-105":29,"detail-sidebar-cat-0-en-105":90},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":11,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},86395,16904993612988,"Olivia Brown","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","StanceMoE: Mixture-of-Experts Architecture for Stance Detection","Actor-level stance detection identifies an author’s expressed position toward specific geopolitical actors referenced or implied in text. While transformer models like BERT perform well, they often use unified representations that inadequately capture heterogeneous signals such as contrastive discourse structures, framing cues, and salient lexical indicators. StanceMoE proposes a context-enhanced Mixture-of-Experts architecture with a fine-tuned BERT encoder, six expert modules, and a context-aware gating mechanism that adaptively routes contributions. Experiments on StanceNakba 2026 Subtask A achieve a macro-F1 of 94.26%, exceeding baselines and BERT variants.","arXiv :2604 .00878v2 [ cs .CL] 13 Jul 2026  \nStanceMoE: Mixture-of-Experts Architecture for Stance Detection  \nAbdullah Al Shafi, Md. Milon Islam, Sk. Imran Hossain, K. M. Azharul Hasan  \nDepartment of Computer Science and Engineering, Khulna University of Engineering & Technology [abdullah@iict.kuet.ac.bd](abdullah@iict.kuet.ac.bd), {milonislam, imran, [az}@cse.kuet.ac.bd](az}@cse.kuet.ac.bd)  \nAbstract  \nActor-level stance detection aims to determine an author’s expressed position toward specific geopolitical actors mentioned or implicated in a text. Although transformer-based models have achieved relatively good performance in stance classification, they typically rely on unified representations that may not sufficiently capture heterogeneous linguistic signals, such as contrastive discourse structures, framing cues, and salient lexical indicators. This motivates the need for adaptive architectures that explicitly model diverse stance-expressive patterns. In this paper, we propose StanceMoE, a context-enhanced Mixture-of-Experts (MoE) architecture built upon a fine-tuned BERT encoder for actor-level stance detection. Our model integrates six expert modules designed to capture complementary linguistic signals, including global semantic orientation, salient lexical cues, clause-level focus, phrase-level patterns, framing indicators, and contrast-driven discourse shifts. A context-aware gating mechanism dynamically weights expert contributions, enabling adaptive routing based on input characteristics. Experiments are conducted on the StanceNakba 2026 Subtask A dataset, comprising 1,401 annotated English texts where the target actor is implicit in the text. StanceMoE achieves a macro-F1 score of 94 .26%, outperforming traditional baselines, and alternative BERT-based variants.  \nKeywords: stance detection, mixture-of-experts, context-aware gating, adaptive weighting.  \n1. Introduction  \nStance detection refers to the automatic identification of an author’s position towards a specific topic, entity, or proposition expressed within a text (Garg and Caragea, 2024) . The task aims to identify whether the presented viewpoint is supportive, opposing, or neutral toward the specific target. The study of stance detection differs from common sentiment analysis because it needs to measure emotional expressions that depend on particular targets, whereas the same text can show different stances based on which target the reader chooses to consider (Niu et al. , 2024) . The nature of this task requires target awareness because it generates both semantic complexity and computational difficulties, particularly when targets remain unmentioned and stances are expressed in an indirect way (Küçük and Can , 2020) .  \nTo address these challenges, we introduce StanceMoE1 , a context-enhanced Mixtureof-Experts (MoE) architecture designed for actor-level stance detection. Unlike conventional transformer-based approaches that rely on single aggregated representations, our method explicitly decomposes stance modeling into complementary expert modules that capture diverse linguistic and discourse-level phenomena. By incorporating a context-aware gating mechanism, the architecture enables adaptive and input-sensitive fusion of heterogeneous stance signals. Through compre-  \n1 [https://github.com/AbdullahRatulk/](https://github.com/AbdullahRatulk/)[ ](https://github.com/AbdullahRatulk/)StanceMoE  \nhensive experimentation and detailed ablation analysis on StanceNakba 2026 Shared Task dataset (Aldous et al. , 2026), we demonstrate that explicitly modeling such diverse patterns enables more robust and fine-grained actor-level stance detection with 3rd position in the competition.  \n2. Related Works  \nThe field of stance detection has evolved through its transition from rule-based systems (Küçük and Can, 2020) to classical supervised systems (Alturayeif et al. , 2023), which rely on manually created features. Deep learning methods, such as Convolutional Neural Net","cbCaircfXbiMye25","https://ap.wps.com/l/cbCaircfXbiMye25","pdf",1345024,3,1,"English","en",105,"# Introduction\n# Related Works\n# Proposed StanceMoE Architecture\n## Contextual Encoder\n## Expert Modules","[{\"question\":\"What does actor-level stance detection aim to determine?\",\"answer\":\"It determines an author’s expressed position toward specific geopolitical actors that are mentioned or implicated in a text.\"},{\"question\":\"Why are transformer baselines not sufficient for this task?\",\"answer\":\"They often rely on unified representations that may fail to capture heterogeneous stance signals, including contrastive discourse structures, framing cues, and salient lexical indicators.\"},{\"question\":\"How does StanceMoE improve stance detection performance?\",\"answer\":\"It combines a fine-tuned BERT encoder with six complementary expert modules and a context-aware gating mechanism that dynamically weights expert contributions, enabling adaptive routing based on the input.\"}]",1784211478,20,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":27},"stancemoe-mixture-of-experts-architecture-for-stance-detection","",{"@graph":35,"@context":84},[36,52,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,49],{"item":40,"name":41,"@type":42,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":20},"https://docshare.wps.com/document/research-report/",{"item":50,"name":13,"@type":42,"position":51},"https://docshare.wps.com/document/stancemoe-mixture-of-experts-architecture-for-stance-detection/86395/",4,{"url":50,"name":13,"@type":53,"author":54,"headline":13,"publisher":56,"fileFormat":59,"inLanguage":23,"description":14,"dateModified":60,"datePublished":61,"encodingFormat":59,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":55},"Person",{"url":40,"name":57,"@type":58},"DocShare","Organization","application/pdf","2026-07-27","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":20},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"What does actor-level stance detection aim to determine?","Question",{"text":74,"@type":75},"It determines an author’s expressed position toward specific geopolitical actors that are mentioned or implicated in a text.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"Why are transformer baselines not sufficient for this task?",{"text":79,"@type":75},"They often rely on unified representations that may fail to capture heterogeneous stance signals, including contrastive discourse structures, framing cues, and salient lexical indicators.",{"name":81,"@type":72,"acceptedAnswer":82},"How does StanceMoE improve stance detection performance?",{"text":83,"@type":75},"It combines a fine-tuned BERT encoder with six complementary expert modules and a context-aware gating mechanism that dynamically weights expert contributions, enabling adaptive routing based on the input.","https://schema.org",{"og:url":50,"og:type":86,"og:title":13,"og:site_name":57,"og:description":14},"article",{"robots":88,"canonical":50},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,119,122,126,129,133],{"id":21,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":51,"doc_module":4,"doc_module_name":45,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":28,"slug":125},9,"Religion & Spirituality","religion-spirituality",{"id":28,"doc_module":4,"doc_module_name":45,"category_name":127,"show_sort_weight":28,"slug":128},"World Cup","world-cup",{"id":130,"doc_module":4,"doc_module_name":45,"category_name":131,"show_sort_weight":130,"slug":132},10,"Lifestyle","lifestyle",{"id":134,"doc_module":4,"doc_module_name":45,"category_name":135,"show_sort_weight":105,"slug":136},19,"General","general"]